AI News #125: Week Ending February 20, 2026 with 49 Executive Summaries
About This Week’s Covers
This week’s cover is inspired by a short film produced by one of China’s top movie directors, Jia Zhangke, using ByteDance’s Seedance 2.0. It’s a five-minute video that is almost impossible to tell is AI (Spoiler alert: he purposefully made a few bits look like AI, because they “play AI in the movie”).
Jia Zhangke is considered one of the greatest living directors… not some AI hack. He’s worth Googling.
I’m overwhelmed by both the critical acclaim he has, and his brazen “look at me make an AI video” flex. I’ll put more about the video in the newsletter itself, as it is the top story.
For this week’s cover images, I took an AI summary of Jia’s aesthetics and combined it with an article about the Chinese Year of the Fire Horse, because this week also celebrated the Chinese New Year. I ran the two through my Claude Python script preparer, and man, did it nail it. At times, the presence of a drab Jia-inspired horse is pretty hilarious.
The script did a great job keeping the priority on Jia’s style, followed by a simple interpretation of the category, the inclusion of the horse, and then the bold title text, which is amazingly translated into Chinese. I spot-checked at least half of the images, and everything seems accurate in Chinese!
For Safe Supercomputing Inc, it swapped the horse with a person wearing a horse mask trying to sneak into the line…
This week marks the first time I’ve been affected by the sheer power of AI transcending the traditional concept of slop. Slop implies ease of use, and low quality and disposable content.
We have photorealistic slop with verisimilitude and mise-en-scène (blasphemy!) that can be created in a matter of minutes.
I made 55 cover images in less than 5 minutes using a Python script that operated without me at the computer. And even though it’s slop, the results are effective as category covers, and yes…they are creative.
I did not ask it or prompt it for any of the creative details.
Interestingly, human creativity benchmarks have been a historic challenge, and it’s an interesting juxtaposition against AI benchmarks.
In my case, I took the OAD test, as well as the Strength Finder test while working in management at American Eagle Outfitters’ home office.
At the time, I got the highest creativity score in the history of the company, since American Eagle had used testing. Fight me 🙂
My OAD result. The letters “CR” represent my creativity score was maxed out and actually off the chart.
It’s notable in the context of AI, that the creativity benchmark that I scored so highly on is not a standard definition of creativity.
The way it was explained to me is that if I was given a blank piece of paper and a brick, I could most likely write more ideas for what someone could do with a brick than anyone else in the room. So creativity, in this case, would be more like free association or an ability to make connections.
I respect the insane capability of artificial intelligence to make connections, and I associate this directly with creativity.
It reminds me of the book Where Good Ideas Come From, by Steven Berlin Johnson, and the concept of the adjacent possible.
The Adjacent Possible
Originally proposed by biologist Stuart Kauffman, suggests that progress happens through incremental steps rather than massive, disconnected leaps.
The Room Metaphor: Imagine standing in a room with four doors. Opening one door leads to a new room with its own set of doors. You couldn’t have reached those new doors without first entering the intermediate room.
Spare Parts: Innovation is about taking the “spare parts” (ideas, technologies, or skills) currently available and recombining them in new ways.
The “Ahead of Its Time” Problem: Ideas fail when they try to jump too far beyond the adjacent possible. For example, Charles Babbage designed a computer in the 1800s, but it failed because the “spare parts” (like vacuum tubes or microchips) didn’t exist yet.
“Disneyland is presented as imaginary in order to make us believe that the rest is real, whereas all of Los Angeles and the America that surrounds it are no longer real, but belong to the hyperreal order and to the order of simulation. It is no longer a question of a false representation of reality (ideology) but of concealing the fact that the real is no longer real, and thus of saving the reality principle.” -Jean Baudrillard, Simulacra and Simulation
This week, I organized 602 headlines! 105 of them informed the executive summaries. I’ve organized the summaries alphabetically by company name, with an occasional category thrown in. First, a few stories worth putting at the top:
Top Stories/Favorites
ByteDance Seedance 2.0 – Video Turing Test Passed – In my opinion
I am not sure if I’ll ever be able to know if a video is real moving forward. I think this marks an important milestone.
the first official AI movie is here and.. its wild. China’s top director Jia Zhangke was so impressed by Seedance 2.0 that he made a film himself… in just 3 days. when asked if AI will replace filmmakers, he said cinema has always moved with tech. Digital cameras didn’t kill film. AI will just make it faster, simpler and better https://x.com/EHuanglu/status/2023449238114320514
Jia Zhangke is considered one of the greatest living directors… not an AI hack. He’s worth Googling.
Jia’s creative ability to stitch together clips and compositions using Seedance has finally broken the short 10-second clip mark that was plaguing slop for so long.
I’m overwhelmed by his critical acclaim, and his brazen “look at me make an AI video” flex.
Here are some links to learn more about Jia’s career and body of work:
Here’s Jia’s video. As you watch this video, look at the absolutely crazy level of detail across every bit of its composition, and remember that not a single thing is real. It’s 100% AI generated. Also, read the Variety article snippet below the clip.
“Hello everyone, for this year’s Chinese New Year, I collaborated with Seedance 2.0, a video generation model developed by Doubao, to create this somewhat unique short film, ‘Jia Zhangke’s Dance,’” Jia posted on social media.
“As the producer, I did not act in the film. The two “Jia Zhangkes” on screen were both generated by Seedance 2.0. We gave one a distinct “AI feel,” while the other is almost exactly like me in real life. Seeing the two “Jia Zhangkes” conversing on screen evokes a strange sense of time travel. From black and white to color, from silent to sound, from film to digital, cinema has changed through years. Every change has been accompanied by doubt and unease. But the development of AI has been so rapid, from being instantly identifiable as fake a few years ago to today’s ability to generate a fairly well-made video from a single sentence – it’s truly fast.”
From Variety:
In the short film, the interaction begins with Jia expressing surprise at finding himself replaced by an AI double during a shoot. The synthetic version explains that it has enhanced his appearance by removing wrinkles and reducing his weight, prompting the director to joke that he wants the missing pounds restored because the altered version looks awkward.
The two then debate whether the AI should be considered a creative work or merely a high-quality imitation. To demonstrate its capabilities, the AI visually transports Jia through a series of shifting cinematic landscapes, placing the director inside stylized environments that evoke the visual worlds associated with his films.
A central conflict emerges when the AI inserts an optimistic line about looking toward a new era, which Jia objects to, saying his characters have never spoken in such terms. The AI counters that once a work reaches audiences, its interpretation no longer belongs solely to its creator.
The conversation also explores the prospect of human-AI collaboration, with the AI proposing a division of labor in which the filmmaker provides ideas while the machine supplies computational power. Jia responds with a joke about his lifelong dislike of “Party A” – Chinese industry slang for clients – leading to a punchline about becoming what one once opposed.
The film ultimately reveals the entire scenario to be a staged performance, with actors discussing the difficulty of portraying Jia Zhangke and suggesting that embodying the director is less about physical likeness than capturing a particular mental state. The video ends with both Jia and his AI counterpart delivering a Lunar New Year greeting.
Seedance 2.0 is ByteDance’s AI video-generation model capable of producing cinematic clips from text, image and audio inputs while maintaining character consistency across scenes. The technology has drawn growing attention across the global film industry alongside criticism from studios and trade groups over alleged copyright violations and unauthorized use of intellectual property and performer likenesses.
Jia has previously spoken about artificial intelligence’s role in filmmaking. During a Venice Film Festival masterclass last year, he said: “AI feels like playing chess at home, while shooting with a camera is like climbing a mountain outdoors. Different directors will choose different tools, but I’m still drawn to the camera and the real world.”
Best known for socially grounded works including “Still Life” and “A Touch of Sin,” Jia has long explored the social and technological transformations of contemporary China.
OpenAI Buys OpenClaw There will be more on this next week, but the headline is the important part.
Excited to work with Peter Steinberger to build the future of agents for everyone and to continue to improve Codex in leaps and bounds. We are committed to OSS, continuing to make OpenClaw flourish and bringing agents to life in a way that is fun, safe and highly productive. https://x.com/thsottiaux/status/2023147973421785386
Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings.
OpenClaw will live in a foundation as an open source project that OpenAI will continue to support. The future is going to be extremely multi-agent and it’s important to us to support open source as part of that. https://x.com/sama/status/2023150230905159801
AI Agents Crushing Benchmarks We estimate that Claude Opus 4.6 has a 50%-time-horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks. While this is the highest point estimate we’ve reported, this measurement is extremely noisy because our current task suite is nearly saturated. https://x.com/METR_Evals/status/2024923422867030027
Agent time on task doubles every five months This is a new separate estimate for LLM time horizon doubling times and it mostly agrees with METR In this case ~4.8-5.7 months https://x.com/scaling01/status/2023350946139435357
GPT-5.2 derived a novel result in theoretical physics, showing that a type of particle interaction many physicists expected would not occur can in fact arise under specific conditions. There is great promise in the potential of AI to benefit people by accelerating science. https://x.com/gdb/status/2022394113971360145
GPT 5.2 derived a new result in theoretical physics. For decades it’s been assumed that certain gluon amplitudes (“”single minus””) were zero, and that the maximally helicity violating amplitudes had two gluons of one helicity and n-2 of the other. It turns out that isn’t https://x.com/kevinweil/status/2022388305434939693
There have been fair questions on whether LLM contributions to STEM are overhyped, but I’ve spoken with physicists about this result and they’ve told me it is a truly significant research contribution, roughly at the level of a solid journal paper, and GPT-5.2 played a key role. https://x.com/polynoamial/status/2022413904757035167
I spent last night with Andrew Strominger and Alex Lupsasca, two of the top physicists in the world They just released a paper, co-authored with OpenAi, that seems to me like ASI Andrew, who helped develop string theory, told me that a year ago, his view was that he didn’t know https://x.com/patrick_oshag/status/2022395157648195801
Anthropic is prepared to loosen its current terms of use, but wants to ensure its tools aren’t used to spy on Americans en masse, or to develop weapons that fire with no human involvement. The Pentagon has aid, that Anthropic will “”pay a price”” for that behavior. Within this https://x.com/kimmonismus/status/2023419652378955809
NEW: Pentagon is so furious with Anthropic for insisting on limiting use of AI for domestic surveillance + autonomous weapons they’re threatening to label the company a “supply chain risk,” forcing vendors to cut ties. https://x.com/DavidLawler10/status/2023425130148626767
Job Risk Former Google AI leader Jad Tarifi warns that long degrees like law, medicine, and even PhDs may become outdated before students graduate, as AI rapidly reaches PhD-level performance. With 70% (!) of AI PhDs now heading into private sector jobs (up https://x.com/kimmonismus/status/2023446044873560178
I think the chance of mass unemployment* from AI is overrated in 2 years and underrated in 7**. Same is true for many effects of AI. * Putting aside gov jobs programs, people specifically wanting to employ a human, etc. ** Among SF/AI-ish people. https://x.com/RyanPGreenblatt/status/2023219133916332070
We are pleased to announce the close of Thrive X. Exceeding $10 billion, Thrive X comprises $1 billion designated for early-stage investments and $9 billion designated for growth-stage investments. We do not view this as a milestone, but as a commitment to the long work ahead. https://x.com/JoshuaKushner/status/2023732796649271619
Anthropic
Sonnet 4.6 Note: Anthropic has an interesting and consistent naming convention that is worth understanding if you don’t know it already: Haiku is their smallest model, Sonnet is their average model, and Opus is the flagship.
The general trend is that Opus sets the lead, and then a few months later, Sonnet comes out with a model that is equally powerful as the previous Opus. During that time, the new Opus is being trained. Meanwhile, Haiku stays lean and mean as a cheap model that improves in the background, for everyday operational tasks.
This can be counterintuitive because, for short periods of time, the latest Sonnet will be tied with the latest version of Opus.
However, Sonnet will be orders of magnitude cheaper and faster. Then the pattern continues, and a few weeks or months later, a new Opus is released that becomes the biggest and best. It’s important to remember this because sometimes you’ll hear Sonnet breaking records, and you’ll wonder what happened to Opus. It’s simply that ebb and flow of the two models as they leapfrog each other.
Claude Sonnet 4.6 is the new leader in GDPval-AA, slightly ahead of Anthropic’s Opus 4.6 on agentic performance of real-world knowledge work tasks less than two weeks after its launch In our pre-release testing with @AnthropicAI, Sonnet 4.6 reached an ELO of 1633 using the https://x.com/ArtificialAnlys/status/2023821893846135212
NEW: Anthropic releases Claude Sonnet 4.6 Nears Opus-level performance across coding and reasoning at Sonnet pricing ($3/$15 per mil tokens). Computer use scores have gone from single digits last year to 72.5% now 📈 + a 1M token context window https://x.com/TheRundownAI/status/2023821446380978238
Most agent actions on our API are low risk. 73% of tool calls appear to have a human in the loop, and only 0.8% are irreversible. But at the frontier, we see agents acting on security systems, financial transactions, and production deployments (though some may be evals). https://x.com/AnthropicAI/status/2024210050718585017
New Anthropic research: Measuring AI agent autonomy in practice. We analyzed millions of interactions across Claude Code and our API to understand how much autonomy people grant to agents, where they’re deployed, and what risks they may pose. Read more: https://x.com/AnthropicAI/status/2024210035480678724
Software engineering makes up ~50% of agentic tool calls on our API, but we see emerging use in other industries. As the frontier of risk and autonomy expands, post-deployment monitoring becomes essential. We encourage other model developers to extend this research. https://x.com/AnthropicAI/status/2024210053369385192
Something strange is happening with AI agents that this new Anthropic research quietly surfaces. The agents are asking us for help more than we’re stepping in to correct *them*. Anthropic analyzed data from Claude Code and their public API to measure how autonomous AI agents https://x.com/omarsar0/status/2024864635120451588
Revenues OpenAI may be a household name, but Anthropic could soon be earning more revenue. Since each company hit $1B in annualized revenues, Anthropic has grown substantially faster (10× vs 3.4× per year) and could overtake OpenAI by mid-2026 if recent trends continue. https://x.com/EpochAIResearch/status/2024536468618956868
Opus For Security Auditing OpenSource Opus4.6 found 500+ vulnerabilities in open-source code and we’ve begun reporting them and contributing patches quick excerpts from some of them 🧵 https://x.com/trq212/status/2024937919937741290
Claude’s Constitution (Very Long) People should read the Claude Constitution. It does a pretty good job of laying out what Anthropic presumably really believes (and it is part of training). I’d think that a clear debate over things that are good or bad or missing there would be helpful. https://x.com/emollick/status/2023612474474303530
Deeper Excel Integration For Claude in Excel users, our add-in now supports MCP connectors, letting Claude work with tools like S&P Global, LSEG, Daloopa, PitchBook, Moody’s and FactSet. Pull in context from outside your spreadsheet without ever leaving Excel. https://x.com/claudeai/status/2023817143096406246
Future of Work To get an idea of the near-term future of work with AI, take a look at the official Claude Cowork plugins, which give the AI specialized knowledge for various hard tasks A natural successor for GPTs, but built for agents (& therefore much more scalable & customizable for firms) https://x.com/emollick/status/2023113346162336137
BitDance is our lunar new year gift: a 14B parameters autoregressive image generation model – that is autoregressive in bits, not codebooks It’s fast given 14B parameters, try it: https://x.com/multimodalart/status/2023797260057014372
ByteDance releases BitDance: Scaling Autoregressive Generative Models with Binary Tokens “”We present BitDance, a scalable autoregressive (AR) image generator that predicts binary visual tokens instead of codebook indices. https://x.com/iScienceLuvr/status/2023707945104458097
Excited to launch Gemini 3.1 Pro! Major improvements across the board including in core reasoning and problem solving. For example scoring 77.1% on the ARC-AGI-2 benchmark – more than 2x the performance of 3 Pro. Rolling out today in @GeminiApp, @antigravity and more – enjoy! https://x.com/demishassabis/status/2024519780976177645
Gemini 3.1 Pro is here! It’s top 3 across Text and Vision Arena, and #6 in Code Arena, tied closely with Claude Opus 4.5. Highlights: ▪️Tied #1 in Text (scoring 1500), 4 pts from Opus 4.6 ▪️Top 3 in Arena Expert Leaderboard (scoring 1538), just behind Opus 4.6 ▪️#6 in Code https://x.com/arena/status/2024519891295089063
Gemini 3.1 Pro is here. Hitting 77.1% on ARC-AGI-2, it’s a step forward in core reasoning (more than 2x 3 Pro). With a more capable baseline, it’s great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative https://x.com/sundarpichai/status/2024516418855981298
Gemini 3.1 Pro landed today. This is based on the same model behind the agentic DeepThink released last week; it is now available to all Gemini users on many apps. This is a really good model especially in reasoning and multimodal understanding/generation. Try it out. https://x.com/mirrokni/status/2024525808501477568
Multimodal function calling is now available in the Gemini Interactions API, build agents that can see and process images natively. 🖼️ Tools return actual images, not text descriptions 👁️ Gemini 3 natively processes returned images 🛠️ Function results support mixed text and https://x.com/_philschmid/status/2022349886318928158
To the Scientist, the Engineer, and the Developer: Gemini 3.1 Pro has arrived in @GeminiApp It’s a significant leap in complex reasoning (77.1% on ARC-AGI-2) so it’s great at agentic tasks, intricate coding, and data synthesis projects. You should see fewer errors, better https://x.com/joshwoodward/status/2024515741819842623
Today, we’re continuing to push the boundaries of AI with our release of Gemini 3.1 Pro. This updated model scores 77.1% on ARC-AGI-2, more than double the reasoning performance of its predecessor, Gemini 3 Pro. Check out the visible improvement in this side-by-side comparison, https://x.com/JeffDean/status/2024525132266688757
Update regarding Gemini 3.1 Pro: -Ranked #1 among all Gemini models released to date. -Ranked #1 among all models I have tested so far. (GPT-5.2 high 165.9 vs Gemini 3.1 Pro 166.6) https://x.com/Hangsiin/status/2024605310913216614
Introducing Lyria 3, our latest and most advanced music model, available in the Gemini App starting today : ) Go from idea, image, or video to music in seconds! https://x.com/OfficialLoganK/status/2024153948488118513 Meet Lyria 3, our latest music generation model from @GoogleDeepMind. 🎶 Now, you can create custom music tracks in the @GeminiApp — just by describing an idea or uploading an image or video. https://x.com/Google/status/2024154379838705920
We just dropped Lyria 3: our latest generative music model. 🔊 It can turn photos and text into dynamic tracks – complete with vocals and lyrics. 🧵 https://x.com/GoogleDeepMind/status/2024153067654902014 We just launched Lyria 3! Our most advanced AI music model in the @GeminiApp 🎵 – Generates 30-second tracks from text or image prompts. – Support custom lyrics, vocals, and cover art. – Supports 8 languages including English, Japanese, and Korean. – All outputs watermarked https://x.com/_philschmid/status/2024154542061805988
Watermarks
All tracks generated in Gemini are embedded with SynthID, our imperceptible watermark for identifying Google AI-generated content. We are also giving you more tools to help identify AI content, broadening our verification capabilities to include audio. Simply upload a file and https://x.com/GeminiApp/status/2024153548641177781 Is that track AI-generated? Now you can just ask @GeminiApp. We’ve broadened our verification tools so you can now upload audio files to Gemini to check for SynthID — our imperceptible watermark on AI-generated content. Just upload a file and ask: “”Was this created using Google “ https://x.com/Google/status/2024172104711823678
Trade Secret Theft News Alert: Today, the #FBI arrested three Silicon Valley engineers who are facing charges of conspiring to commit trade secret theft from Google and other leading technology companies, theft and attempted theft of trade secrets, and obstruction of justice. https://x.com/FBISanFrancisco/status/2024670479974363376
MiniMax 2.5 I’ve been maintaining this newsletter for 125 weeks, and I try not to create a new category for specific companies or models unless they really merit one. Minimax has been appearing quite a bit in recent weeks, but I have not yet made a category for it.
However, I highly recommend familiarizing yourself with Minimax, and if they keep up this pace, I’ll have to add a category soon.
How efficient is MiniMax M2.5? We benchmarked on 8xH200 TEP8 with @vllm_project . At a reasonable 10-25s TTFT, M2.5 is able to sustain ~2500 tok/s/GPU throughput. For decode, it’s still possible to reach ~20 tok/s/GPU throughput at a strict 20 tok/s/user interactivity with 10K+ https://x.com/SemiAnalysis_/status/2023418414203646066 MLX
MiniMax 2.5 running LOCALLY on a single M3 Ultra 512GB! Writing a poem on LLMs at 6bit quantization! 🔥 Let’s start some coding, context and distributed tests! Generation: 40.2 tokens-per-sec Peak memory: 186 GB https://x.com/ivanfioravanti/status/2022338870172684655
Alignment We’re committing $7.5M to @AISecurityInst’s Alignment Project to fund independent research on mitigations for safety and security risks from misaligned AI. https://x.com/OpenAINewsroom/status/2024546609485533442
🎉 Congrats to @Alibaba_Qwen on releasing Qwen3.5 on Chinese New Year’s Eve — day-0 support is ready in vLLM! Qwen3.5 is a multimodal MoE with Gated Delta Networks architecture — 397B total params, only 17B active. What makes it interesting for inference: 🧠 Gated Delta https://x.com/vllm_project/status/2023341059343061138
🔥 Alibaba’s Qwen 3.5 just dropped — and Zhihu is dissecting it. Here’s a sharp breakdown from Zhihu contributor toyama nao 👇 🏆 Verdict: “”The spearhead of the open-source elite.”” 📊 Big picture Tongyi Lab’s pattern: new mid-size model leapfrogs old giant. • Last cycle: 80B https://x.com/ZhihuFrontier/status/2024176484232155236
Alibaba’s new Qwen3.5-397B-A17B is the #3 open weights model in the Artificial Analysis Intelligence Index – a significant upgrade from Qwen3-235B-A22B-2507, and achieved with fewer active parameters than leading peers Qwen3.5-397B-A17B is the first model released by Alibaba https://x.com/ArtificialAnlys/status/2023794497055060262
Qwen3.5 is Live! Today we openweight the first model, Qwen3-397B-A17B, which is a native multimodal model supporting both thinking and non-thinking modes. We have strengthened its coding and agentic capabilities to foster productivity for developers and enterprises. Hope you https://x.com/JustinLin610/status/2023332446713070039
You can now run Qwen3.5 locally! 💜 Qwen3.5-397B-A17B is an open MoE vision reasoning LLM for agentic coding & chat. It performs on par with Gemini 3 Pro, Claude Opus 4.5 & GPT-5.2. Run 4-bit on 256GB Mac / RAM. Guide: https://x.com/UnslothAI/status/2023338222601064463
Simulations and Vision Pixels are all you need! Just kidding 🙂 Whether or not explicit 3D representations survive the bitter lesson, one thing is pretty clear — vision & robotics, perceiving & acting are on a glorious collision course. https://x.com/bilawalsidhu/status/2023902733632208938
Full Executive Summaries with Links, Generated by Claude Sonnet 4.5
ByteDance’s Seedance 2.0 enables acclaimed director Jia Zhangke to create film in three days Renowned Chinese filmmaker Jia Zhangke used ByteDance’s new Seedance 2.0 AI video generator to produce a complete short film for Chinese New Year in just three days, marking a significant milestone as one of cinema’s most respected auteurs embraces AI filmmaking tools. This represents a major validation for AI video technology, as Jia Zhangke is considered among China’s greatest living directors with decades of critical acclaim, suggesting AI video generation has reached professional filmmaking standards rather than remaining a novelty for casual creators.
the first official AI movie is here and.. its wild China’s top director Jia Zhangke was so impressed by Seedance 2.0 that he made a film himself.. in just 3 days when asked if AI will replace filmmakers, he said cinema has always moved with tech. Digital cameras didn’t kill https://x.com/EHuanglu/status/2023449238114320514
Jia Zhangke is considered one of the greatest living directors… not an AI hack. He’s worth Googling. I’m overwhelmed by the critical acclaim he has, and his brazen “look at me make an AI video” flex. The X video has subtitles, and the YouTube one has commentary in the description from Vanity Fair.
OpenAI acquires viral AI agent OpenClaw creator in major strategic shift OpenAI hired Peter Steinberger, creator of the open-source OpenClaw agent that gained explosive adoption among developers for autonomously completing tasks across applications and systems. The acquisition signals the industry’s pivot from conversational AI to autonomous agents that can browse, code, and execute tasks independently. OpenClaw’s viral success came from its “unhinged” approach with minimal safeguards—capabilities that major labs typically can’t release due to security concerns.
Claude Opus 4.6 can complete software tasks in 14.5 hours on average Anthropic’s latest AI model shows dramatically faster software development capabilities, though the measurement is imprecise because current benchmarks are reaching their limits. This represents a significant leap in AI’s ability to handle complex programming work, suggesting we may need entirely new ways to measure AI performance as models approach human-level coding speeds.
We estimate that Claude Opus 4.6 has a 50%-time-horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks. While this is the highest point estimate we’ve reported, this measurement is extremely noisy because our current task suite is nearly saturated. https://x.com/METR_Evals/status/2024923422867030027
AI model capabilities are doubling every 5 months, new analysis confirms Independent research validates that large language models are improving at twice the rate previously estimated, with capabilities doubling roughly every five months rather than annually. This accelerated timeline suggests AI systems may reach critical benchmarks much sooner than anticipated, potentially compressing the window for developing safety measures and regulatory frameworks.
OpenAI’s GPT-5.2 discovers new theoretical physics result about particle interactions GPT-5.2 derived a novel finding about gluon particle interactions that physicists had assumed were impossible for decades, co-authoring a paper with leading researchers Andrew Strominger and Alex Lupsasca. The discovery overturns long-held assumptions about specific particle scattering amplitudes and represents what physicists describe as a significant research contribution equivalent to a solid journal publication. This marks a notable shift from AI assisting research to AI making original scientific discoveries that advance human knowledge in fundamental physics.
GPT-5.2 derived a novel result in theoretical physics, showing that a type of particle interaction many physicists expected would not occur can in fact arise under specific conditions. There is great promise in the potential of AI to benefit people by accelerating science. https://x.com/gdb/status/2022394113971360145
GPT 5.2 derived a new result in theoretical physics. For decades it’s been assumed that certain gluon amplitudes (“”single minus””) were zero, and that the maximally helicity violating amplitudes had two gluons of one helicity and n-2 of the other. It turns out that isn’t https://x.com/kevinweil/status/2022388305434939693
There have been fair questions on whether LLM contributions to STEM are overhyped, but I’ve spoken with physicists about this result and they’ve told me it is a truly significant research contribution, roughly at the level of a solid journal paper, and GPT-5.2 played a key role. https://x.com/polynoamial/status/2022413904757035167
I spent last night with Andrew Strominger and Alex Lupsasca, two of the top physicists in the world They just released a paper, co-authored with OpenAi, that seems to me like ASI Andrew, who helped develop string theory, told me that a year ago, his view was that he didn’t know https://x.com/patrick_oshag/status/2022395157648195801
Pentagon threatens to label Anthropic supply chain risk over AI limits The Pentagon is threatening to designate Anthropic as a “supply chain risk” after the AI company insisted on restricting use of its Claude chatbot for domestic surveillance and fully autonomous weapons. This escalation follows reports that the military already used Claude in operations against Venezuelan leader Maduro, highlighting growing tensions between AI companies’ safety guardrails and government demands for unrestricted access. The designation would force other vendors to cut ties with Anthropic, potentially isolating one of the leading AI safety-focused companies from government contracts.
Anthropic is prepared to loosen its current terms of use, but wants to ensure its tools aren’t used to spy on Americans en masse, or to develop weapons that fire with no human involvement. The Pentagon has aid, that Anthropic will “”pay a price”” for that behavior. Within this https://x.com/kimmonismus/status/2023419652378955809
NEW: Pentagon is so furious with Anthropic for insisting on limiting use of AI for domestic surveillance + autonomous weapons they’re threatening to label the company a “supply chain risk,” forcing vendors to cut ties. With @m_ccuri and @mikeallen https://x.com/DavidLawler10/status/2023425130148626767
Fei-Fei Li’s World Labs raises $1 billion for spatial intelligence AI The startup, led by AI pioneer Fei-Fei Li, secured funding from AMD, Nvidia, and Autodesk to develop AI systems that understand and generate 3D worlds rather than just processing flat images or text. This represents a major bet on “spatial intelligence” as the next frontier in AI, with applications spanning robotics, virtual reality, and scientific discovery. World Labs’ Marble product already lets users create persistent 3D environments from simple text or image prompts.
Former Google AI leader warns degrees may become obsolete before graduation Jad Tarifi argues that AI’s rapid advancement to PhD-level performance could make traditional long-term degrees in law, medicine, and academia irrelevant by the time students complete them. The shift is already visible as 70% of AI PhDs now choose private sector jobs over academia, suggesting the market is already recognizing this transformation.
AI Makes Degrees Obsolete Former Google AI leader Jad Tarifi warns that long degrees like law, medicine, and even PhDs may become outdated before students graduate, as AI rapidly reaches PhD-level performance. With 70% (!) of AI PhDs now heading into private sector jobs (up https://x.com/kimmonismus/status/2023446044873560178
AI unemployment fears misplaced for near term, underestimated long term A prominent AI researcher argues that while mass job displacement from AI is unlikely within two years, the seven-year outlook poses greater risks than many in Silicon Valley currently anticipate. The timing distinction suggests AI’s labor market disruption will be gradual rather than immediate, giving society more time to adapt but potentially creating complacency about longer-term preparation needs.
I think the chance of mass unemployment* from AI is overrated in 2 years and underrated in 7**. Same is true for many effects of AI. * Putting aside gov jobs programs, people specifically wanting to employ a human, etc. ** Among SF/AI-ish people. https://x.com/RyanPGreenblatt/status/2023219133916332070
Saudi Arabia invests $3 billion in Musk’s xAI startup Saudi Arabia’s state-backed Humain became a major shareholder in xAI through a $20 billion funding round, just before SpaceX acquired the AI company for $1.25 trillion. This deal reflects Gulf nations’ aggressive push into AI as they diversify from oil, with sovereign wealth funds now controlling over $4 trillion and making strategic bets across the industry. The investment gives xAI a key customer for its Grok chatbot while strengthening Musk’s ties to the kingdom.
AI startups raised $6 billion in mega-rounds during January 2026 alone Seventeen US AI companies secured funding rounds of $100 million or more in just the first two months of 2026, with Anthropic leading at a $380 billion valuation and SkildAI raising $1.4 billion for robotics AI. This pace suggests 2026 could surpass 2025’s record $76 billion in AI mega-rounds, with investors particularly focused on specialized applications like voice AI, robotics, and medical chatbots rather than general-purpose models.
Thrive Capital raises $10 billion fund focused on AI investments The venture firm closed one of the largest tech funds ever, with $9 billion earmarked for growth-stage AI companies and $1 billion for early startups. This massive capital pool signals institutional confidence in AI’s commercial potential and could accelerate the development of AI applications across industries. The fund size reflects investor belief that AI will generate returns comparable to previous tech booms like mobile and cloud computing.
We are pleased to announce the close of Thrive X. Exceeding $10 billion, Thrive X comprises $1 billion designated for early-stage investments and $9 billion designated for growth-stage investments. We do not view this as a milestone, but as a commitment to the long work ahead. https://x.com/JoshuaKushner/status/2023732796649271619
New arena reveals AI models struggle with real-world spreadsheet creation Researchers tested leading AI models from OpenAI, Anthropic, and Google on spreadsheet generation, finding that formatting and visual presentation matter more to users than complex formulas. Even top models fail to follow professional conventions like financial color-coding, with expert evaluations agreeing with crowd preferences only half the time. The findings suggest current AI benchmarks miss crucial real-world usability factors that determine whether business users actually find AI-generated spreadsheets useful.
Stanford researchers create AI that simulates users by modeling internal mental states HumanLM goes beyond copying writing styles to generate underlying psychological states like stance and emotion before crafting responses, using reinforcement learning to align these internal states with how real users actually behave. This approach could enable more realistic user testing and personalized AI applications by capturing the deeper reasoning behind human responses rather than just surface-level language patterns.
FAIL 🙂 I don’t see any content provided about “The Simulation Company” to summarize. Could you please share the actual news material or article content you’d like me to turn into the two-line executive summary format? I need the specific details about what happened, the business implications, and any relevant evidence to create the factual headline and explanatory paragraph you’re looking for. FAIL 🙂
Anthropic’s Claude Sonnet 4.6 delivers near-flagship performance at mid-tier pricing The new model matches Opus-level capabilities on coding and reasoning tasks while maintaining Sonnet’s $3-15 per million token pricing, with computer use skills jumping to 72.5% accuracy. Users preferred it over the previous flagship Opus 4.5 model 59% of the time, suggesting Anthropic has compressed high-end AI performance into a more affordable package.
Claude Sonnet 4.6 is the new leader in GDPval-AA, slightly ahead of Anthropic’s Opus 4.6 on agentic performance of real-world knowledge work tasks less than two weeks after its launch In our pre-release testing with @AnthropicAI, Sonnet 4.6 reached an ELO of 1633 using the https://x.com/ArtificialAnlys/status/2023821893846135212
NEW: Anthropic releases Claude Sonnet 4.6 Nears Opus-level performance across coding and reasoning at Sonnet pricing ($3/$15 per mil tokens). Computer use scores have gone from single digits last year to 72.5% now 📈 + a 1M token context window https://x.com/TheRundownAI/status/2023821446380978238
Anthropic study reveals AI agents now work 45 minutes autonomously Anthropic analyzed millions of real-world interactions and found that experienced users let AI agents work twice as long without interruption compared to three months ago, reaching over 45 minutes for the longest sessions. The research shows agents pause to ask humans for help more often than humans interrupt them, and while most agent actions remain low-risk software tasks, emerging use in healthcare, finance, and cybersecurity signals a need for new oversight frameworks as autonomy expands.
Most agent actions on our API are low risk. 73% of tool calls appear to have a human in the loop, and only 0.8% are irreversible. But at the frontier, we see agents acting on security systems, financial transactions, and production deployments (though some may be evals). https://x.com/AnthropicAI/status/2024210050718585017
New Anthropic research: Measuring AI agent autonomy in practice. We analyzed millions of interactions across Claude Code and our API to understand how much autonomy people grant to agents, where they’re deployed, and what risks they may pose. Read more: https://x.com/AnthropicAI/status/2024210035480678724
Software engineering makes up ~50% of agentic tool calls on our API, but we see emerging use in other industries. As the frontier of risk and autonomy expands, post-deployment monitoring becomes essential. We encourage other model developers to extend this research. https://x.com/AnthropicAI/status/2024210053369385192
Something strange is happening with AI agents that this new Anthropic research quietly surfaces. The agents are asking us for help more than we’re stepping in to correct *them*. Anthropic analyzed data from Claude Code and their public API to measure how autonomous AI agents https://x.com/omarsar0/status/2024864635120451588
Claude now writes code to filter web search results before processing them Anthropic’s latest Claude models can automatically generate and run code to filter irrelevant web content during searches, improving accuracy by 11% while using 24% fewer tokens. This addresses a key inefficiency where AI agents previously had to process entire web pages to find relevant information. The feature works by letting Claude dynamically parse and filter search results before loading them into its context window, making web-based research tasks more precise and cost-effective.
Anthropic CEO predicts AI geniuses in data centers within years Dario Amodei maintains his timeline for extremely rapid AI progress, expecting “country of geniuses in a datacenter” capabilities within a decade and possibly this year, while Anthropic’s revenue exploded from zero to $10 billion in just three years. His predictions stand out for their aggressive speed—claiming 90% of code will be AI-written within months—though he acknowledges the gap between raw capabilities and real-world adoption remains a major constraint. The interview notably downplayed existential risks and alignment concerns despite the unprecedented pace of development.
Anthropic’s revenue growth rate triples OpenAI’s despite smaller size Since both companies reached $1 billion in annual revenue, Anthropic has grown 10 times per year compared to OpenAI’s 3.4 times, positioning the Claude maker to potentially surpass ChatGPT’s creator by mid-2026. This suggests the AI market remains highly competitive with room for multiple winners, challenging assumptions about OpenAI’s dominance in commercial AI applications.
OpenAI may be a household name, but Anthropic could soon be earning more revenue. Since each company hit $1B in annualized revenues, Anthropic has grown substantially faster (10× vs 3.4× per year) and could overtake OpenAI by mid-2026 if recent trends continue. https://x.com/EpochAIResearch/status/2024536468618956868
AI system discovers over 500 security flaws in open-source software Opus4.6’s automated vulnerability detection represents a significant leap in AI-powered code security, potentially transforming how the software industry identifies and fixes critical flaws that could affect millions of users. The system is already contributing patches back to projects, demonstrating AI’s growing capability to not just find problems but actively help solve them.
Opus4.6 found 500+ vulnerabilities in open-source code and we’ve begun reporting them and contributing patches quick excerpts from some of them 🧵 https://x.com/trq212/status/2024937919937741290
Anthropic’s Claude Constitution reveals company’s core AI values and training principles The publicly available document outlines Anthropic’s fundamental beliefs about AI behavior and safety, offering transparency into how Claude is trained and providing a concrete framework for public debate about AI ethics and governance.
People should read the Claude Constitution. It does a pretty good job of laying out what Anthropic presumably really believes (and it is part of training). I’d think that a clear debate over things that are good or bad or missing there would be helpful. https://x.com/emollick/status/2023612474474303530
Claude Excel add-in now connects to major financial data providers Anthropic’s spreadsheet tool can now pull live data from S&P Global, Bloomberg’s LSEG, and other premium financial services directly into Excel workflows. This bridges AI assistance with institutional-grade data sources that typically require separate subscriptions and manual data transfers, potentially streamlining financial analysis for professionals who rely on both Excel and these specialized databases.
For Claude in Excel users, our add-in now supports MCP connectors, letting Claude work with tools like S&P Global, LSEG, Daloopa, PitchBook, Moody’s and FactSet. Pull in context from outside your spreadsheet without ever leaving Excel. https://x.com/claudeai/status/2023817143096406246
I don’t see any AI news content in your message to summarize. You’ve included what appears to be a partial quote about forbidding something on third-party open source code, but there’s no complete article or news item provided. Could you please share the full AI news content you’d like me to summarize into the two-line executive summary format?
Claude launches workplace plugins for specialized professional tasks Anthropic has released official plugins that give Claude AI specialized knowledge for complex work functions, representing a shift from general chatbots to task-specific AI agents. Unlike OpenAI’s GPTs which are primarily conversational, these plugins are designed for scalable deployment across organizations. This signals the evolution toward AI systems that can handle specialized professional workflows rather than just answering questions.
To get an idea of the near-term future of work with AI, take a look at the official Claude Cowork plugins, which give the AI specialized knowledge for various hard tasks A natural successor for GPTs, but built for agents (& therefore much more scalable & customizable for firms) https://x.com/emollick/status/2023113346162336137
Anthropic study finds AI coding assistance reduces skill mastery by 17% Software developers using AI assistance scored significantly lower on coding comprehension tests compared to those coding manually, despite minor speed gains. The randomized trial of 52 engineers revealed that while AI can accelerate task completion, it often leads to “cognitive offloading” that undermines learning—particularly debugging skills crucial for overseeing AI-generated code. However, developers who used AI strategically by asking follow-up questions and seeking explanations maintained better skill retention, suggesting the interaction approach matters more than AI use itself.
ByteDance launches BitDance, a 14 billion parameter image generator using binary tokens BitDance generates images by predicting binary code directly rather than using traditional codebook methods, potentially offering faster processing speeds. This approach represents a technical shift in how AI models create images, moving away from the discrete token systems used by most current generators. The model’s 14 billion parameters make it substantial but not record-breaking, with ByteDance positioning speed as its key advantage over existing image generation systems.
BitDance is our lunar new year gift: a 14B parameters autoregressive image generation model – that is autoregressive in bits, not codebooks It’s fast given 14B parameters, try it: https://x.com/multimodalart/status/2023797260057014372
ByteDance releases BitDance: Scaling Autoregressive Generative Models with Binary Tokens “”We present BitDance, a scalable autoregressive (AR) image generator that predicts binary visual tokens instead of codebook indices. With high-entropy binary latents, BitDance lets each https://x.com/iScienceLuvr/status/2023707945104458097
ElevenLabs launches specialized AI voice platform for government agencies The AI voice synthesis company has created a dedicated government version of its platform, offering enhanced security controls and compliance features for public sector use. This marks a significant shift as AI voice technology moves from consumer entertainment into official government communications, potentially transforming how agencies interact with citizens while raising new questions about authenticity in public discourse.
ElevenLabs secures first-of-its-kind insurance coverage for AI voice agents The voice AI company partnered with Lloyd’s of London to create unprecedented insurance protecting against potential harms from AI-generated speech, including deepfakes and misinformation. This marks the insurance industry’s first attempt to quantify and cover risks specific to AI agents that can speak and interact autonomously. The move signals growing corporate recognition that AI liability extends beyond traditional software failures to include reputational damage and societal harm from synthetic media.
Google launches Gemini 3.1 Pro with major reasoning breakthrough Google’s new Gemini 3.1 Pro AI model scored 77.1% on the ARC-AGI-2 reasoning benchmark, more than doubling its predecessor’s performance and ranking among the top AI models globally. The upgrade represents a significant leap in complex problem-solving abilities, now available across Google’s consumer apps, developer tools, and enterprise platforms. Early testing suggests it may outperform competing models like GPT-5.2 in certain reasoning tasks.
Excited to launch Gemini 3.1 Pro! Major improvements across the board including in core reasoning and problem solving. For example scoring 77.1% on the ARC-AGI-2 benchmark – more than 2x the performance of 3 Pro. Rolling out today in @GeminiApp, @antigravity and more – enjoy! https://x.com/demishassabis/status/2024519780976177645
Gemini 3.1 Pro is here! It’s top 3 across Text and Vision Arena, and #6 in Code Arena, tied closely with Claude Opus 4.5. Highlights: ▪️Tied #1 in Text (scoring 1500), 4 pts from Opus 4.6 ▪️Top 3 in Arena Expert Leaderboard (scoring 1538), just behind Opus 4.6 ▪️#6 in Code https://x.com/arena/status/2024519891295089063
Gemini 3.1 Pro is here. Hitting 77.1% on ARC-AGI-2, it’s a step forward in core reasoning (more than 2x 3 Pro). With a more capable baseline, it’s great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative https://x.com/sundarpichai/status/2024516418855981298
Gemini 3.1 Pro landed today. This is based on the same model behind the agentic DeepThink released last week; it is now available to all Gemini users on many apps. This is a really good model especially in reasoning and multimodal understanding/generation. Try it out. https://x.com/mirrokni/status/2024525808501477568
Multimodal function calling is now available in the Gemini Interactions API, build agents that can see and process images natively. 🖼️ Tools return actual images, not text descriptions 👁️ Gemini 3 natively processes returned images 🛠️ Function results support mixed text and https://x.com/_philschmid/status/2022349886318928158
To the Scientist, the Engineer, and the Developer: Gemini 3.1 Pro has arrived in @GeminiApp It’s a significant leap in complex reasoning (77.1% on ARC-AGI-2) so it’s great at agentic tasks, intricate coding, and data synthesis projects. You should see fewer errors, better https://x.com/joshwoodward/status/2024515741819842623
Today, we’re continuing to push the boundaries of AI with our release of Gemini 3.1 Pro. This updated model scores 77.1% on ARC-AGI-2, more than double the reasoning performance of its predecessor, Gemini 3 Pro. Check out the visible improvement in this side-by-side comparison, https://x.com/JeffDean/status/2024525132266688757
Update regarding Gemini 3.1 Pro: -Ranked #1 among all Gemini models released to date. -Ranked #1 among all models I have tested so far. (GPT-5.2 high 165.9 vs Gemini 3.1 Pro 166.6) However, please note that my testing has limitations due to budget constraints: -I have not https://x.com/Hangsiin/status/2024605310913216614
Google launches Lyria 3 music generator in Gemini app Google’s new Lyria 3 AI model lets users create 30-second music tracks from text descriptions or uploaded photos directly in the Gemini app. Unlike previous music AI tools, Lyria 3 automatically generates lyrics, offers creative control over style and tempo, and produces more musically complex compositions. The feature includes built-in copyright protections and SynthID watermarking to identify AI-generated content.
Meet Lyria 3, our latest music generation model from @GoogleDeepMind. 🎶 Now, you can create custom music tracks in the @GeminiApp — just by describing an idea or uploading an image or video. https://x.com/Google/status/2024154379838705920
We just launched Lyria 3! Our most advanced AI music model in the @GeminiApp 🎵 – Generates 30-second tracks from text or image prompts. – Support custom lyrics, vocals, and cover art. – Supports 8 languages including English, Japanese, and Korean. – All outputs watermarked with https://x.com/_philschmid/status/2024154542061805988
Google launches audio watermark detection tool in Gemini app Users can now upload audio files to Gemini to check for Google’s SynthID watermark, expanding AI content verification beyond text and images to help identify artificially generated music and speech.
All tracks generated in Gemini are embedded with SynthID, our imperceptible watermark for identifying Google AI-generated content. We are also giving you more tools to help identify AI content, broadening our verification capabilities to include audio. Simply upload a file and https://x.com/GeminiApp/status/2024153548641177781
Is that track AI-generated? Now you can just ask @GeminiApp. We’ve broadened our verification tools so you can now upload audio files to Gemini to check for SynthID — our imperceptible watermark on AI-generated content. Just upload a file and ask: “”Was this created using Google https://x.com/Google/status/2024172104711823678
FBI arrests three Silicon Valley engineers for stealing AI trade secrets Federal agents charged three engineers with conspiring to steal proprietary technology from Google and other major tech companies, marking a significant crackdown on intellectual property theft in the AI sector. The arrests highlight growing concerns about trade secret protection as AI capabilities become increasingly valuable corporate assets, with the FBI treating such theft as a national security priority requiring criminal prosecution.
News Alert: Today, the #FBI arrested three Silicon Valley engineers who are facing charges of conspiring to commit trade secret theft from Google and other leading technology companies, theft and attempted theft of trade secrets, and obstruction of justice. Samaneh Ghandali, 41, https://x.com/FBISanFrancisco/status/2024670479974363376
Meta signs massive Nvidia deal for millions of AI chips Meta will deploy millions of Nvidia chips including new standalone CPUs across 30 planned U.S. data centers, marking the first large-scale use of Nvidia’s Grace processors as standalone units rather than paired with GPUs. The multiyear agreement, valued in the tens of billions, represents a significant expansion beyond Meta’s traditional GPU purchases and supports CEO Mark Zuckerberg’s goal to deliver “personal superintelligence” globally. The deal secures Meta’s access to both current Blackwell and next-generation Rubin chips amid industry-wide supply constraints.
Manus launches AI agents that work inside Telegram messaging app Manus released AI agents that integrate directly into Telegram, allowing users to access full AI capabilities through simple chat messages without complex setup or maintenance. This matters because it removes technical barriers that typically prevent widespread adoption of personal AI agents, making advanced multi-step reasoning and task execution as accessible as sending a text message. The service supports voice, images, and files while maintaining the same core capabilities as Manus’s web platform.
MiniMax’s M2.5 model runs efficiently on both server chips and Apple hardware The Chinese AI company’s latest model achieves 2,500 tokens per second on server GPUs and runs locally on Apple’s M3 Ultra at 40 tokens per second, demonstrating unusually broad hardware compatibility. This flexibility could reduce deployment costs and enable more diverse AI applications compared to models that require specific chip architectures.
How efficient is MiniMax M2.5? We benchmarked on 8xH200 TEP8 with @vllm_project . At a reasonable 10-25s TTFT, M2.5 is able to sustain ~2500 tok/s/GPU throughput. For decode, it’s still possible to reach ~20 tok/s/GPU throughput at a strict 20 tok/s/user interactivity with 10K+ https://x.com/SemiAnalysis_/status/2023418414203646066
MLX MiniMax 2.5 running LOCALLY on a single M3 Ultra 512GB! Writing a poem on LLMs at 6bit quantization! 🔥 Let’s start some coding, context and distributed tests! Generation: 40.2 tokens-per-sec Peak memory: 186 GB https://x.com/ivanfioravanti/status/2022338870172684655
OpenAI’s valuation could reach $100 billion in new funding round The ChatGPT maker is reportedly raising capital at a valuation that would make it one of the world’s most valuable private companies, reflecting massive investor confidence in generative AI despite the technology still being in early commercial stages. This represents a dramatic jump from its $29 billion valuation just 10 months ago, signaling that investors believe AI chatbots and similar tools will generate enormous revenues across industries.
Peter Steinberger joins OpenAI to develop next-generation personal agents OpenAI hired the experienced developer to advance AI agents that can interact with each other and perform complex tasks for users, signaling the company’s push beyond chatbots into autonomous digital assistants that could reshape how people work with technology.
Excited to work with Peter Steinberger to build the future of agents for everyone and to continue to improve Codex in leaps and bounds. We are committed to OSS, continuing to make OpenClaw flourish and bringing agents to life in a way that is fun, safe and highly productive. https://x.com/thsottiaux/status/2023147973421785386
Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our https://x.com/sama/status/2023150230905159801
OpenAI hires Instagram’s global partnerships VP Charles Porch OpenAI recruited Charles Porch from Instagram’s global partnerships team, signaling the company’s focus on expanding business relationships beyond pure AI development. This hire suggests OpenAI is prioritizing commercial partnerships and platform integrations as it scales its AI services to reach broader markets.
OpenAI launches benchmark to test AI agents on smart contract security EVMbench specifically measures whether AI systems can find, exploit, and fix serious vulnerabilities in blockchain smart contracts—the self-executing programs that power decentralized finance. This represents a shift toward testing AI on specialized cybersecurity tasks rather than general reasoning, potentially accelerating AI-powered security auditing in the $200 billion DeFi ecosystem.
Introducing EVMbench—a new benchmark that measures how well AI agents can detect, exploit, and patch high-severity smart contract vulnerabilities. https://x.com/OpenAI/status/2024193883748651102
OpenAI launches new safety controls to restrict ChatGPT misuse OpenAI introduced “Lockdown Mode” and “Elevated Risk” labels for ChatGPT to prevent harmful uses like generating illegal content or misinformation. The features automatically detect and block risky requests while flagging potentially dangerous conversations for human review. This represents OpenAI’s most aggressive content moderation system yet, responding to growing regulatory pressure and safety concerns about AI chatbots being weaponized for harmful purposes.
OpenAI commits $7.5 million to independent AI safety research OpenAI is funding external researchers at the AI Security Institute to study how to prevent AI systems from behaving in unintended or harmful ways. This marks a significant shift toward supporting independent safety research rather than keeping all alignment work in-house, potentially addressing concerns about self-regulation in the AI industry.
OpenAI expands beyond software with planned smart speaker and glasses The ChatGPT maker is developing its first consumer hardware devices, marking a strategic shift from pure software to compete directly with Amazon’s Alexa and Meta’s smart glasses in the growing AI-powered device market.
Alibaba releases Qwen 3.5 with 397B parameters but only 17B active The Chinese tech giant’s new AI model ranks third globally among open-source models despite using fewer active parameters than competitors, demonstrating a more efficient architecture. This “mixture of experts” approach activates only a fraction of its total capacity during inference, potentially reducing computational costs while maintaining high performance.
🎉 Congrats to @Alibaba_Qwen on releasing Qwen3.5 on Chinese New Year’s Eve — day-0 support is ready in vLLM! Qwen3.5 is a multimodal MoE with Gated Delta Networks architecture — 397B total params, only 17B active. What makes it interesting for inference: 🧠 Gated Delta https://x.com/vllm_project/status/2023341059343061138
🔥 Alibaba’s Qwen 3.5 just dropped — and Zhihu is dissecting it. Here’s a sharp breakdown from Zhihu contributor toyama nao 👇 🏆 Verdict: “”The spearhead of the open-source elite.”” 📊 Big picture Tongyi Lab’s pattern: new mid-size model leapfrogs old giant. • Last cycle: 80B https://x.com/ZhihuFrontier/status/2024176484232155236
Alibaba’s new Qwen3.5-397B-A17B is the #3 open weights model in the Artificial Analysis Intelligence Index – a significant upgrade from Qwen3-235B-A22B-2507, and achieved with fewer active parameters than leading peers Qwen3.5-397B-A17B is the first model released by Alibaba https://x.com/ArtificialAnlys/status/2023794497055060262
Alibaba releases Qwen3.5, a 397-billion parameter AI model with reasoning capabilities The open-source model matches performance of top proprietary systems like GPT and Claude while running locally on high-end consumer hardware. Its distinctive “thinking mode” allows the AI to show its reasoning process, and the model combines vision, coding, and chat abilities in a single system that developers can download and modify freely.
Qwen3.5 is Live! Today we openweight the first model, Qwen3-397B-A17B, which is a native multimodal model supporting both thinking and non-thinking modes. We have strengthened its coding and agentic capabilities to foster productivity for developers and enterprises. Hope you https://x.com/JustinLin610/status/2023332446713070039
You can now run Qwen3.5 locally! 💜 Qwen3.5-397B-A17B is an open MoE vision reasoning LLM for agentic coding & chat. It performs on par with Gemini 3 Pro, Claude Opus 4.5 & GPT-5.2. Run 4-bit on 256GB Mac / RAM. Guide: https://t.co/wjS1lMnbNp GGUF: https://x.com/UnslothAI/status/2023338222601064463
Reddit tests AI shopping feature that turns user recommendations into purchase links Reddit is piloting an AI-powered search tool that converts community product discussions into interactive shopping carousels with pricing and retailer links. This marks a significant shift from Reddit’s traditional text-based format to direct commerce integration, potentially transforming how social platforms monetize user-generated content. The test targets electronics queries for select U.S. users, surfacing products mentioned in actual Reddit conversations alongside partner retailer catalogs.
I don’t see any specific AI news content to summarize in your message. The text appears to be a casual comment about vision and robotics converging, but lacks the concrete details needed for a factual executive summary – no specific companies, research findings, product launches, or measurable developments are mentioned. Could you please provide the actual AI news items you’d like me to summarize? nan
Pixels are all you need! Just kidding 🙂 Whether or not explicit 3D representations survive the bitter lesson, one thing is pretty clear — vision & robotics, perceiving & acting are on a glorious collision course. https://x.com/bilawalsidhu/status/2023902733632208938
Spotify’s top developers stopped coding in December, CEO reveals Spotify CEO Daniel Ek disclosed that the company’s most senior developers have shifted away from hands-on programming, likely focusing instead on AI-assisted development and higher-level architecture work. This represents a significant change in how even elite tech companies are restructuring engineering roles, suggesting AI tools have advanced enough to handle routine coding tasks that previously required top talent.
WordPress launches AI assistant directly integrated into its editor platform WordPress.com now offers an AI assistant built into its editor and media library for Business and Commerce plan users at no extra cost, allowing site owners to modify layouts, generate content, and create images through conversational commands without leaving their workspace. This marks a shift from standalone AI tools toward integrated assistance that understands existing site context and can take immediate action within the content management system.
Leave a Reply